Microsoft built a fake marketplace to test AI agents — they failed in surprising ways

Finally—someone publicly admitting that AI agents aren't ready. Microsoft's marketplace simulation should've been easy. Instead, agents got confused, made bad decisions, got stuck in loops. That's actual important data.

Look, the hype cycle needs a real failure. This is it. The agents didn't scale to unsupervised tasks. Autonomy isn't just better LLMs—it's a different problem entirely. This sets back the "agentic future" narrative by 18 months, minimum.